My paper on #chatgpt and the Divergent Association Task is now available, and downloadble here: https://lnkd.in/gqM3T2hD
Key conclusions:
1. On the DAT, mean scores for GPT3.5 and GPT4 were higher than the human norm (but, although statistically significant, the effect sizes of these differences were small).
2. However, GPT3.5 and GPT4 have a tendency to repeat certain favorite words, and have a high variance (sometimes they score very high, but sometimes they score very low).
3. This means that, while GPT3.5 and GPT4 are, strictly numerically, "better at" verbal divergent production than humans, that is an oversimplification.
4. While divergent thinking is an important element of creativity, it is not the entirety of creativity. So just being good at divergent thinking doesn't necessarily mean that ChatGPT is "more creative than humans". #creativity #ai